Fixing arm compilation.

[Layer3]

[Layer3]

Sorry for taking so long to follow up on all of this.

I pushed a CI chance to your branch to build wheels and run test on ARM architectures on Linux and MacOS (if MacOS fails, we can just do Linux for now and that's fine).

This shouldn't change anything on Mac anymore, but not entirely sure. I'd leave this completely up to you. I don't own a mac platform, though I might get one if the number of projects requiring mac support continues following this trend for me.

Taking a closer look at simde, it seems there is no performance overhead in just using it directly. I wonder if we could skip the NATIVE_SSE flag and just always use simde. I guess we'll need to check where else uses SSE instructions.

Definitely! The big problem with this imo is that it definitely requires more setup and testing, but the gain is there. As far as I've seen this requires some more involved changes to the simd abstraction in geogram. I'm not opposed to haivng a look at this, just saying that it is definitely a bit of work.

One more point about the NATIVE_SSE flag. Currently this feels more like a BUILD_KHADAS flag that has the added benefit of having a high chance to work on other platforms that might have problems to compile with sse. The actual issue we are having is that Khadas-Linux isn't playing perfectly nice with CMake I think. Also in Geogram this might be resolved differently, by adding an additional platform for arm Linux and make sure that arm boards switch to that configuration instead of defaulting to linux-gcc-dynamic.

Sorry for the novel, hope I'm not stating the obvious too often. : )

[fwilliams]

Okay I think I fixed the CI and merged these changes. I'll try and get #55 working on top of these changes!

[Layer3]

Cool, thank you!